diff mbox series

RDMA/device: Fix a race between mad_client and cm_client init

Message ID 20240102034335.34842-1-lishifeng@sangfor.com.cn (mailing list archive)
State Superseded
Headers show
Series RDMA/device: Fix a race between mad_client and cm_client init | expand

Commit Message

Shifeng Li Jan. 2, 2024, 3:43 a.m. UTC
The mad_client will be initialized in enable_device_and_get(), while the
devices_rwsem will be downgraded to a read semaphore. There is a window
that leads to the failed initialization for cm_client, since it can not
get matched mad port from ib_mad_port_list, and the matched mad port will
be added to the list after that.

    mad_client    |                       cm_client
------------------|--------------------------------------------------------
ib_register_device|
enable_device_and_get
down_write(&devices_rwsem)
xa_set_mark(&devices, DEVICE_REGISTERED)
downgrade_write(&devices_rwsem)
                  |
                  |ib_cm_init
                  |ib_register_client(&cm_client)
                  |down_read(&devices_rwsem)
                  |xa_for_each_marked (&devices, DEVICE_REGISTERED)
                  |add_client_context
                  |cm_add_one
                  |ib_register_mad_agent
                  |ib_get_mad_port
                  |__ib_get_mad_port
                  |list_for_each_entry(entry, &ib_mad_port_list, port_list)
                  |return NULL
                  |up_read(&devices_rwsem)
                  |
add_client_context|
ib_mad_init_device|
ib_mad_port_open  |
list_add_tail(&port_priv->port_list, &ib_mad_port_list)
up_read(&devices_rwsem)
                  |

Fix it by using the devices_rwsem write semaphore to protect the mad_client
init flow in enable_device_and_get().

Fixes: d0899892edd0 ("RDMA/device: Provide APIs from the core code to help unregistration")
Cc: Shifeng Li <lishifeng1992@126.com>
Signed-off-by: Shifeng Li <lishifeng@sangfor.com.cn>
---
 drivers/infiniband/core/device.c | 8 +-------
 1 file changed, 1 insertion(+), 7 deletions(-)

Comments

Leon Romanovsky Jan. 2, 2024, 8:58 a.m. UTC | #1
On Mon, Jan 01, 2024 at 07:43:35PM -0800, Shifeng Li wrote:
> The mad_client will be initialized in enable_device_and_get(), while the
> devices_rwsem will be downgraded to a read semaphore. There is a window
> that leads to the failed initialization for cm_client, since it can not
> get matched mad port from ib_mad_port_list, and the matched mad port will
> be added to the list after that.
> 
>     mad_client    |                       cm_client
> ------------------|--------------------------------------------------------
> ib_register_device|
> enable_device_and_get
> down_write(&devices_rwsem)
> xa_set_mark(&devices, DEVICE_REGISTERED)
> downgrade_write(&devices_rwsem)
>                   |
>                   |ib_cm_init
>                   |ib_register_client(&cm_client)
>                   |down_read(&devices_rwsem)
>                   |xa_for_each_marked (&devices, DEVICE_REGISTERED)
>                   |add_client_context
>                   |cm_add_one
>                   |ib_register_mad_agent
>                   |ib_get_mad_port
>                   |__ib_get_mad_port
>                   |list_for_each_entry(entry, &ib_mad_port_list, port_list)
>                   |return NULL
>                   |up_read(&devices_rwsem)
>                   |
> add_client_context|
> ib_mad_init_device|
> ib_mad_port_open  |
> list_add_tail(&port_priv->port_list, &ib_mad_port_list)
> up_read(&devices_rwsem)
>                   |

How is this stack possible?

ib_register_device() is called by drivers and happens much later than ib_cm_init().

Thanks

> 
> Fix it by using the devices_rwsem write semaphore to protect the mad_client
> init flow in enable_device_and_get().
> 
> Fixes: d0899892edd0 ("RDMA/device: Provide APIs from the core code to help unregistration")
> Cc: Shifeng Li <lishifeng1992@126.com>
> Signed-off-by: Shifeng Li <lishifeng@sangfor.com.cn>
> ---
>  drivers/infiniband/core/device.c | 8 +-------
>  1 file changed, 1 insertion(+), 7 deletions(-)
> 
> diff --git a/drivers/infiniband/core/device.c b/drivers/infiniband/core/device.c
> index 67bcea7a153c..85782786993d 100644
> --- a/drivers/infiniband/core/device.c
> +++ b/drivers/infiniband/core/device.c
> @@ -1315,12 +1315,6 @@ static int enable_device_and_get(struct ib_device *device)
>  	down_write(&devices_rwsem);
>  	xa_set_mark(&devices, device->index, DEVICE_REGISTERED);
>  
> -	/*
> -	 * By using downgrade_write() we ensure that no other thread can clear
> -	 * DEVICE_REGISTERED while we are completing the client setup.
> -	 */
> -	downgrade_write(&devices_rwsem);
> -
>  	if (device->ops.enable_driver) {
>  		ret = device->ops.enable_driver(device);
>  		if (ret)
> @@ -1337,7 +1331,7 @@ static int enable_device_and_get(struct ib_device *device)
>  	if (!ret)
>  		ret = add_compat_devs(device);
>  out:
> -	up_read(&devices_rwsem);
> +	up_write(&devices_rwsem);
>  	return ret;
>  }
>  
> -- 
> 2.25.1
> 
>
Shifeng Li Jan. 2, 2024, 11:33 a.m. UTC | #2
On 2024/1/2 16:58, Leon Romanovsky wrote:
> On Mon, Jan 01, 2024 at 07:43:35PM -0800, Shifeng Li wrote:
>> The mad_client will be initialized in enable_device_and_get(), while the
>> devices_rwsem will be downgraded to a read semaphore. There is a window
>> that leads to the failed initialization for cm_client, since it can not
>> get matched mad port from ib_mad_port_list, and the matched mad port will
>> be added to the list after that.
>>
>>      mad_client    |                       cm_client
>> ------------------|--------------------------------------------------------
>> ib_register_device|
>> enable_device_and_get
>> down_write(&devices_rwsem)
>> xa_set_mark(&devices, DEVICE_REGISTERED)
>> downgrade_write(&devices_rwsem)
>>                    |
>>                    |ib_cm_init
>>                    |ib_register_client(&cm_client)
>>                    |down_read(&devices_rwsem)
>>                    |xa_for_each_marked (&devices, DEVICE_REGISTERED)
>>                    |add_client_context
>>                    |cm_add_one
>>                    |ib_register_mad_agent
>>                    |ib_get_mad_port
>>                    |__ib_get_mad_port
>>                    |list_for_each_entry(entry, &ib_mad_port_list, port_list)
>>                    |return NULL
>>                    |up_read(&devices_rwsem)
>>                    |
>> add_client_context|
>> ib_mad_init_device|
>> ib_mad_port_open  |
>> list_add_tail(&port_priv->port_list, &ib_mad_port_list)
>> up_read(&devices_rwsem)
>>                    |
> 
> How is this stack possible?
> 
> ib_register_device() is called by drivers and happens much later than ib_cm_init().
> 

I've caught the stack and err log as follows, ib_mad_init_device() called after
cm_add_one().

[   98.281786] CPU: 18 PID: 30079 Comm: modprobe Kdump: loaded
[   98.281787] Call Trace:
[   98.281790]  dump_stack+0x71/0xab
[   98.281805]  ib_register_mad_agent+0x1c6/0x27f0 [ib_core]
[   98.281840]  cm_add_one+0x4b0/0x9d0 [ib_cm]
[   98.281865]  add_client_context+0x2b9/0x380 [ib_core]
[   98.281890]  ib_register_client+0x22a/0x2a0 [ib_core]
[   98.281908]  __init_backport+0x12f/0x1000 [ib_cm]
[   98.281912]  do_one_initcall+0x87/0x2e2
[   98.281926]  do_init_module+0x1c3/0x5f7
[   98.281928]  load_module+0x4fe0/0x68d0
[   98.281945]  __do_sys_finit_module+0x14d/0x180
[   98.281952]  do_syscall_64+0xa0/0x370
[   98.281957]  entry_SYSCALL_64_after_hwframe+0x65/0xca
[   98.281958] RIP: 0033:0x7f51d3a4373d
[   98.281961] RSP: 002b:00007ffceb925938 EFLAGS: 00000202 ORIG_RAX: 0000000000000139
[   98.281964] RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007f51d3a4373d
[   98.281965] RDX: 0000000000000000 RSI: 0000000001636b70 RDI: 0000000000000003
[   98.281966] RBP: 00007ffceb925950 R08: 0000000000000000 R09: 0000000001636b70
[   98.281967] R10: 0000000000000003 R11: 0000000000000202 R12: 0000000000402410
[   98.281968] R13: 00007ffceb926e60 R14: 0000000000000000 R15: 0000000000000000

[   98.281977] infiniband mlx5_1: ib_register_mad_agent: Invalid port 1

[   98.349040] CPU: 38 PID: 29896 Comm: modprobe Kdump: loaded
[   98.349043] Call Trace:
[   98.349053]  dump_stack+0x71/0xab
[   98.349076]  ib_register_mad_agent+0x1c6/0x27f0 [ib_core]
[   98.349149]  ib_agent_port_open+0xe2/0x2d0 [ib_core]
[   98.349164]  ib_mad_init_device+0x818/0x1d70 [ib_core]
[   98.349197]  add_client_context+0x2b9/0x380 [ib_core]
[   98.349221]  enable_device_and_get+0x1ab/0x340 [ib_core]
[   98.349292]  ib_register_device+0xcbf/0xfd0 [ib_core]
[   98.349355]  __mlx5_ib_add+0x44/0xf0 [mlx5_ib]
[   98.349404]  mlx5_add_device+0xc3/0x280 [mlx5_core]
[   98.349434]  mlx5_register_interface+0x109/0x190 [mlx5_core]
[   98.349442]  do_one_initcall+0x87/0x2e2
[   98.349460]  do_init_module+0x1c3/0x5f7
[   98.349462]  load_module+0x4fe0/0x68d0
[   98.349481]  __do_sys_init_module+0x1f9/0x220
[   98.349486]  do_syscall_64+0xa0/0x370
[   98.349492]  entry_SYSCALL_64_after_hwframe+0x65/0xca
[   98.349494] RIP: 0033:0x7fbc327faa7a
[   98.349500] RSP: 002b:00007fff890df7f8 EFLAGS: 00000202 ORIG_RAX: 00000000000000af
[   98.349502] RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007fbc327faa7a
[   98.349503] RDX: 00000000004250c0 RSI: 00000000001b0230 RDI: 00007fbc31919010
[   98.349505] RBP: 00007fff890df860 R08: 00000000025045b0 R09: 0000000000000000
[   98.349506] R10: 00000000001b0230 R11: 0000000000000202 R12: 0000000000402410
[   98.349507] R13: 00007fff890e0cf0 R14: 0000000000000000 R15: 0000000000000000

Thanks

> Thanks
> 
>>
>> Fix it by using the devices_rwsem write semaphore to protect the mad_client
>> init flow in enable_device_and_get().
>>
>> Fixes: d0899892edd0 ("RDMA/device: Provide APIs from the core code to help unregistration")
>> Cc: Shifeng Li <lishifeng1992@126.com>
>> Signed-off-by: Shifeng Li <lishifeng@sangfor.com.cn>
>> ---
>>   drivers/infiniband/core/device.c | 8 +-------
>>   1 file changed, 1 insertion(+), 7 deletions(-)
>>
>> diff --git a/drivers/infiniband/core/device.c b/drivers/infiniband/core/device.c
>> index 67bcea7a153c..85782786993d 100644
>> --- a/drivers/infiniband/core/device.c
>> +++ b/drivers/infiniband/core/device.c
>> @@ -1315,12 +1315,6 @@ static int enable_device_and_get(struct ib_device *device)
>>   	down_write(&devices_rwsem);
>>   	xa_set_mark(&devices, device->index, DEVICE_REGISTERED);
>>   
>> -	/*
>> -	 * By using downgrade_write() we ensure that no other thread can clear
>> -	 * DEVICE_REGISTERED while we are completing the client setup.
>> -	 */
>> -	downgrade_write(&devices_rwsem);
>> -
>>   	if (device->ops.enable_driver) {
>>   		ret = device->ops.enable_driver(device);
>>   		if (ret)
>> @@ -1337,7 +1331,7 @@ static int enable_device_and_get(struct ib_device *device)
>>   	if (!ret)
>>   		ret = add_compat_devs(device);
>>   out:
>> -	up_read(&devices_rwsem);
>> +	up_write(&devices_rwsem);
>>   	return ret;
>>   }
>>   
>> -- 
>> 2.25.1
>>
>>
>
Leon Romanovsky Jan. 2, 2024, 11:58 a.m. UTC | #3
On Tue, Jan 02, 2024 at 07:33:41PM +0800, Shifeng Li wrote:
> On 2024/1/2 16:58, Leon Romanovsky wrote:
> > On Mon, Jan 01, 2024 at 07:43:35PM -0800, Shifeng Li wrote:
> > > The mad_client will be initialized in enable_device_and_get(), while the
> > > devices_rwsem will be downgraded to a read semaphore. There is a window
> > > that leads to the failed initialization for cm_client, since it can not
> > > get matched mad port from ib_mad_port_list, and the matched mad port will
> > > be added to the list after that.
> > > 
> > >      mad_client    |                       cm_client
> > > ------------------|--------------------------------------------------------
> > > ib_register_device|
> > > enable_device_and_get
> > > down_write(&devices_rwsem)
> > > xa_set_mark(&devices, DEVICE_REGISTERED)
> > > downgrade_write(&devices_rwsem)
> > >                    |
> > >                    |ib_cm_init
> > >                    |ib_register_client(&cm_client)
> > >                    |down_read(&devices_rwsem)
> > >                    |xa_for_each_marked (&devices, DEVICE_REGISTERED)
> > >                    |add_client_context
> > >                    |cm_add_one
> > >                    |ib_register_mad_agent
> > >                    |ib_get_mad_port
> > >                    |__ib_get_mad_port
> > >                    |list_for_each_entry(entry, &ib_mad_port_list, port_list)
> > >                    |return NULL
> > >                    |up_read(&devices_rwsem)
> > >                    |
> > > add_client_context|
> > > ib_mad_init_device|
> > > ib_mad_port_open  |
> > > list_add_tail(&port_priv->port_list, &ib_mad_port_list)
> > > up_read(&devices_rwsem)
> > >                    |
> > 
> > How is this stack possible?
> > 
> > ib_register_device() is called by drivers and happens much later than ib_cm_init().
> > 
> 
> I've caught the stack and err log as follows, ib_mad_init_device() called after


ib_mad_init_device != ib_register_device

You mentioned ib_register_device() in your callstack, while you caught ib_mad_init_device().

Thanks

> cm_add_one().
> 
> [   98.281786] CPU: 18 PID: 30079 Comm: modprobe Kdump: loaded
> [   98.281787] Call Trace:
> [   98.281790]  dump_stack+0x71/0xab
> [   98.281805]  ib_register_mad_agent+0x1c6/0x27f0 [ib_core]
> [   98.281840]  cm_add_one+0x4b0/0x9d0 [ib_cm]
> [   98.281865]  add_client_context+0x2b9/0x380 [ib_core]
> [   98.281890]  ib_register_client+0x22a/0x2a0 [ib_core]
> [   98.281908]  __init_backport+0x12f/0x1000 [ib_cm]
> [   98.281912]  do_one_initcall+0x87/0x2e2
> [   98.281926]  do_init_module+0x1c3/0x5f7
> [   98.281928]  load_module+0x4fe0/0x68d0
> [   98.281945]  __do_sys_finit_module+0x14d/0x180
> [   98.281952]  do_syscall_64+0xa0/0x370
> [   98.281957]  entry_SYSCALL_64_after_hwframe+0x65/0xca
> [   98.281958] RIP: 0033:0x7f51d3a4373d
> [   98.281961] RSP: 002b:00007ffceb925938 EFLAGS: 00000202 ORIG_RAX: 0000000000000139
> [   98.281964] RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007f51d3a4373d
> [   98.281965] RDX: 0000000000000000 RSI: 0000000001636b70 RDI: 0000000000000003
> [   98.281966] RBP: 00007ffceb925950 R08: 0000000000000000 R09: 0000000001636b70
> [   98.281967] R10: 0000000000000003 R11: 0000000000000202 R12: 0000000000402410
> [   98.281968] R13: 00007ffceb926e60 R14: 0000000000000000 R15: 0000000000000000
> 
> [   98.281977] infiniband mlx5_1: ib_register_mad_agent: Invalid port 1
> 
> [   98.349040] CPU: 38 PID: 29896 Comm: modprobe Kdump: loaded
> [   98.349043] Call Trace:
> [   98.349053]  dump_stack+0x71/0xab
> [   98.349076]  ib_register_mad_agent+0x1c6/0x27f0 [ib_core]
> [   98.349149]  ib_agent_port_open+0xe2/0x2d0 [ib_core]
> [   98.349164]  ib_mad_init_device+0x818/0x1d70 [ib_core]
> [   98.349197]  add_client_context+0x2b9/0x380 [ib_core]
> [   98.349221]  enable_device_and_get+0x1ab/0x340 [ib_core]
> [   98.349292]  ib_register_device+0xcbf/0xfd0 [ib_core]
> [   98.349355]  __mlx5_ib_add+0x44/0xf0 [mlx5_ib]
> [   98.349404]  mlx5_add_device+0xc3/0x280 [mlx5_core]
> [   98.349434]  mlx5_register_interface+0x109/0x190 [mlx5_core]
> [   98.349442]  do_one_initcall+0x87/0x2e2
> [   98.349460]  do_init_module+0x1c3/0x5f7
> [   98.349462]  load_module+0x4fe0/0x68d0
> [   98.349481]  __do_sys_init_module+0x1f9/0x220
> [   98.349486]  do_syscall_64+0xa0/0x370
> [   98.349492]  entry_SYSCALL_64_after_hwframe+0x65/0xca
> [   98.349494] RIP: 0033:0x7fbc327faa7a
> [   98.349500] RSP: 002b:00007fff890df7f8 EFLAGS: 00000202 ORIG_RAX: 00000000000000af
> [   98.349502] RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007fbc327faa7a
> [   98.349503] RDX: 00000000004250c0 RSI: 00000000001b0230 RDI: 00007fbc31919010
> [   98.349505] RBP: 00007fff890df860 R08: 00000000025045b0 R09: 0000000000000000
> [   98.349506] R10: 00000000001b0230 R11: 0000000000000202 R12: 0000000000402410
> [   98.349507] R13: 00007fff890e0cf0 R14: 0000000000000000 R15: 0000000000000000
> 
> Thanks
> 
> > Thanks
> > 
> > > 
> > > Fix it by using the devices_rwsem write semaphore to protect the mad_client
> > > init flow in enable_device_and_get().
> > > 
> > > Fixes: d0899892edd0 ("RDMA/device: Provide APIs from the core code to help unregistration")
> > > Cc: Shifeng Li <lishifeng1992@126.com>
> > > Signed-off-by: Shifeng Li <lishifeng@sangfor.com.cn>
> > > ---
> > >   drivers/infiniband/core/device.c | 8 +-------
> > >   1 file changed, 1 insertion(+), 7 deletions(-)
> > > 
> > > diff --git a/drivers/infiniband/core/device.c b/drivers/infiniband/core/device.c
> > > index 67bcea7a153c..85782786993d 100644
> > > --- a/drivers/infiniband/core/device.c
> > > +++ b/drivers/infiniband/core/device.c
> > > @@ -1315,12 +1315,6 @@ static int enable_device_and_get(struct ib_device *device)
> > >   	down_write(&devices_rwsem);
> > >   	xa_set_mark(&devices, device->index, DEVICE_REGISTERED);
> > > -	/*
> > > -	 * By using downgrade_write() we ensure that no other thread can clear
> > > -	 * DEVICE_REGISTERED while we are completing the client setup.
> > > -	 */
> > > -	downgrade_write(&devices_rwsem);
> > > -
> > >   	if (device->ops.enable_driver) {
> > >   		ret = device->ops.enable_driver(device);
> > >   		if (ret)
> > > @@ -1337,7 +1331,7 @@ static int enable_device_and_get(struct ib_device *device)
> > >   	if (!ret)
> > >   		ret = add_compat_devs(device);
> > >   out:
> > > -	up_read(&devices_rwsem);
> > > +	up_write(&devices_rwsem);
> > >   	return ret;
> > >   }
> > > -- 
> > > 2.25.1
> > > 
> > > 
> > 
> 
>
Shifeng Li Jan. 2, 2024, 12:30 p.m. UTC | #4
On 2024/1/2 19:58, Leon Romanovsky wrote:
> On Tue, Jan 02, 2024 at 07:33:41PM +0800, Shifeng Li wrote:
>> On 2024/1/2 16:58, Leon Romanovsky wrote:
>>> On Mon, Jan 01, 2024 at 07:43:35PM -0800, Shifeng Li wrote:
>>>> The mad_client will be initialized in enable_device_and_get(), while the
>>>> devices_rwsem will be downgraded to a read semaphore. There is a window
>>>> that leads to the failed initialization for cm_client, since it can not
>>>> get matched mad port from ib_mad_port_list, and the matched mad port will
>>>> be added to the list after that.
>>>>
>>>>       mad_client    |                       cm_client
>>>> ------------------|--------------------------------------------------------
>>>> ib_register_device|
>>>> enable_device_and_get
>>>> down_write(&devices_rwsem)
>>>> xa_set_mark(&devices, DEVICE_REGISTERED)
>>>> downgrade_write(&devices_rwsem)
>>>>                     |
>>>>                     |ib_cm_init
>>>>                     |ib_register_client(&cm_client)
>>>>                     |down_read(&devices_rwsem)
>>>>                     |xa_for_each_marked (&devices, DEVICE_REGISTERED)
>>>>                     |add_client_context
>>>>                     |cm_add_one
>>>>                     |ib_register_mad_agent
>>>>                     |ib_get_mad_port
>>>>                     |__ib_get_mad_port
>>>>                     |list_for_each_entry(entry, &ib_mad_port_list, port_list)
>>>>                     |return NULL
>>>>                     |up_read(&devices_rwsem)
>>>>                     |
>>>> add_client_context|
>>>> ib_mad_init_device|
>>>> ib_mad_port_open  |
>>>> list_add_tail(&port_priv->port_list, &ib_mad_port_list)
>>>> up_read(&devices_rwsem)
>>>>                     |
>>>
>>> How is this stack possible?
>>>
>>> ib_register_device() is called by drivers and happens much later than ib_cm_init().
>>>
>>
>> I've caught the stack and err log as follows, ib_mad_init_device() called after
> 
> 
> ib_mad_init_device != ib_register_device
> 
> You mentioned ib_register_device() in your callstack, while you caught ib_mad_init_device().
> 

ib_register_device() is called by mlx5_ib driver, and then ib_register_device()
calls ib_mad_init_device().

Thanks

> Thanks
> 
>> cm_add_one().
>>
>> [   98.281786] CPU: 18 PID: 30079 Comm: modprobe Kdump: loaded
>> [   98.281787] Call Trace:
>> [   98.281790]  dump_stack+0x71/0xab
>> [   98.281805]  ib_register_mad_agent+0x1c6/0x27f0 [ib_core]
>> [   98.281840]  cm_add_one+0x4b0/0x9d0 [ib_cm]
>> [   98.281865]  add_client_context+0x2b9/0x380 [ib_core]
>> [   98.281890]  ib_register_client+0x22a/0x2a0 [ib_core]
>> [   98.281908]  __init_backport+0x12f/0x1000 [ib_cm]
>> [   98.281912]  do_one_initcall+0x87/0x2e2
>> [   98.281926]  do_init_module+0x1c3/0x5f7
>> [   98.281928]  load_module+0x4fe0/0x68d0
>> [   98.281945]  __do_sys_finit_module+0x14d/0x180
>> [   98.281952]  do_syscall_64+0xa0/0x370
>> [   98.281957]  entry_SYSCALL_64_after_hwframe+0x65/0xca
>> [   98.281958] RIP: 0033:0x7f51d3a4373d
>> [   98.281961] RSP: 002b:00007ffceb925938 EFLAGS: 00000202 ORIG_RAX: 0000000000000139
>> [   98.281964] RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007f51d3a4373d
>> [   98.281965] RDX: 0000000000000000 RSI: 0000000001636b70 RDI: 0000000000000003
>> [   98.281966] RBP: 00007ffceb925950 R08: 0000000000000000 R09: 0000000001636b70
>> [   98.281967] R10: 0000000000000003 R11: 0000000000000202 R12: 0000000000402410
>> [   98.281968] R13: 00007ffceb926e60 R14: 0000000000000000 R15: 0000000000000000
>>
>> [   98.281977] infiniband mlx5_1: ib_register_mad_agent: Invalid port 1
>>
>> [   98.349040] CPU: 38 PID: 29896 Comm: modprobe Kdump: loaded
>> [   98.349043] Call Trace:
>> [   98.349053]  dump_stack+0x71/0xab
>> [   98.349076]  ib_register_mad_agent+0x1c6/0x27f0 [ib_core]
>> [   98.349149]  ib_agent_port_open+0xe2/0x2d0 [ib_core]
>> [   98.349164]  ib_mad_init_device+0x818/0x1d70 [ib_core]
>> [   98.349197]  add_client_context+0x2b9/0x380 [ib_core]
>> [   98.349221]  enable_device_and_get+0x1ab/0x340 [ib_core]
>> [   98.349292]  ib_register_device+0xcbf/0xfd0 [ib_core]
>> [   98.349355]  __mlx5_ib_add+0x44/0xf0 [mlx5_ib]
>> [   98.349404]  mlx5_add_device+0xc3/0x280 [mlx5_core]
>> [   98.349434]  mlx5_register_interface+0x109/0x190 [mlx5_core]
>> [   98.349442]  do_one_initcall+0x87/0x2e2
>> [   98.349460]  do_init_module+0x1c3/0x5f7
>> [   98.349462]  load_module+0x4fe0/0x68d0
>> [   98.349481]  __do_sys_init_module+0x1f9/0x220
>> [   98.349486]  do_syscall_64+0xa0/0x370
>> [   98.349492]  entry_SYSCALL_64_after_hwframe+0x65/0xca
>> [   98.349494] RIP: 0033:0x7fbc327faa7a
>> [   98.349500] RSP: 002b:00007fff890df7f8 EFLAGS: 00000202 ORIG_RAX: 00000000000000af
>> [   98.349502] RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007fbc327faa7a
>> [   98.349503] RDX: 00000000004250c0 RSI: 00000000001b0230 RDI: 00007fbc31919010
>> [   98.349505] RBP: 00007fff890df860 R08: 00000000025045b0 R09: 0000000000000000
>> [   98.349506] R10: 00000000001b0230 R11: 0000000000000202 R12: 0000000000402410
>> [   98.349507] R13: 00007fff890e0cf0 R14: 0000000000000000 R15: 0000000000000000
>>
>> Thanks
>>
>>> Thanks
>>>
>>>>
>>>> Fix it by using the devices_rwsem write semaphore to protect the mad_client
>>>> init flow in enable_device_and_get().
>>>>
>>>> Fixes: d0899892edd0 ("RDMA/device: Provide APIs from the core code to help unregistration")
>>>> Cc: Shifeng Li <lishifeng1992@126.com>
>>>> Signed-off-by: Shifeng Li <lishifeng@sangfor.com.cn>
>>>> ---
>>>>    drivers/infiniband/core/device.c | 8 +-------
>>>>    1 file changed, 1 insertion(+), 7 deletions(-)
>>>>
>>>> diff --git a/drivers/infiniband/core/device.c b/drivers/infiniband/core/device.c
>>>> index 67bcea7a153c..85782786993d 100644
>>>> --- a/drivers/infiniband/core/device.c
>>>> +++ b/drivers/infiniband/core/device.c
>>>> @@ -1315,12 +1315,6 @@ static int enable_device_and_get(struct ib_device *device)
>>>>    	down_write(&devices_rwsem);
>>>>    	xa_set_mark(&devices, device->index, DEVICE_REGISTERED);
>>>> -	/*
>>>> -	 * By using downgrade_write() we ensure that no other thread can clear
>>>> -	 * DEVICE_REGISTERED while we are completing the client setup.
>>>> -	 */
>>>> -	downgrade_write(&devices_rwsem);
>>>> -
>>>>    	if (device->ops.enable_driver) {
>>>>    		ret = device->ops.enable_driver(device);
>>>>    		if (ret)
>>>> @@ -1337,7 +1331,7 @@ static int enable_device_and_get(struct ib_device *device)
>>>>    	if (!ret)
>>>>    		ret = add_compat_devs(device);
>>>>    out:
>>>> -	up_read(&devices_rwsem);
>>>> +	up_write(&devices_rwsem);
>>>>    	return ret;
>>>>    }
>>>> -- 
>>>> 2.25.1
>>>>
>>>>
>>>
>>
>>
>
Jason Gunthorpe Jan. 3, 2024, 6:48 p.m. UTC | #5
On Mon, Jan 01, 2024 at 07:43:35PM -0800, Shifeng Li wrote:
> diff --git a/drivers/infiniband/core/device.c b/drivers/infiniband/core/device.c
> index 67bcea7a153c..85782786993d 100644
> --- a/drivers/infiniband/core/device.c
> +++ b/drivers/infiniband/core/device.c
> @@ -1315,12 +1315,6 @@ static int enable_device_and_get(struct ib_device *device)
>  	down_write(&devices_rwsem);
>  	xa_set_mark(&devices, device->index, DEVICE_REGISTERED);
>  
> -	/*
> -	 * By using downgrade_write() we ensure that no other thread can clear
> -	 * DEVICE_REGISTERED while we are completing the client setup.
> -	 */
> -	downgrade_write(&devices_rwsem);
> -
>  	if (device->ops.enable_driver) {
>  		ret = device->ops.enable_driver(device);
>  		if (ret)
> @@ -1337,7 +1331,7 @@ static int enable_device_and_get(struct ib_device *device)
>  	if (!ret)
>  		ret = add_compat_devs(device);
>  out:
> -	up_read(&devices_rwsem);
> +	up_write(&devices_rwsem);
>  	return ret;
>  }

I don't think messing with the devices_rwsem here is a great idea, it
would be better to address this on the clients_rwsem side like:

diff --git a/drivers/infiniband/core/device.c b/drivers/infiniband/core/device.c
index 67bcea7a153c6a..b956c9f8e62d34 100644
--- a/drivers/infiniband/core/device.c
+++ b/drivers/infiniband/core/device.c
@@ -1730,7 +1730,7 @@ static int assign_client_id(struct ib_client *client)
 {
 	int ret;
 
-	down_write(&clients_rwsem);
+	lockdep_assert_held(&clients_rwsem);
 	/*
 	 * The add/remove callbacks must be called in FIFO/LIFO order. To
 	 * achieve this we assign client_ids so they are sorted in
@@ -1739,14 +1739,11 @@ static int assign_client_id(struct ib_client *client)
 	client->client_id = highest_client_id;
 	ret = xa_insert(&clients, client->client_id, client, GFP_KERNEL);
 	if (ret)
-		goto out;
+		return ret;
 
 	highest_client_id++;
 	xa_set_mark(&clients, client->client_id, CLIENT_REGISTERED);
-
-out:
-	up_write(&clients_rwsem);
-	return ret;
+	return 0;
 }
 
 static void remove_client_id(struct ib_client *client)
@@ -1776,25 +1773,31 @@ int ib_register_client(struct ib_client *client)
 {
 	struct ib_device *device;
 	unsigned long index;
+	bool need_unreg = false;
 	int ret;
 
 	refcount_set(&client->uses, 1);
 	init_completion(&client->uses_zero);
-	ret = assign_client_id(client);
-	if (ret)
-		return ret;
 
 	down_read(&devices_rwsem);
+	down_write(&clients_rwsem);
+	ret = assign_client_id(client);
+	if (ret)
+		goto out;
+
+	need_unreg = true;
 	xa_for_each_marked (&devices, index, device, DEVICE_REGISTERED) {
 		ret = add_client_context(device, client);
-		if (ret) {
-			up_read(&devices_rwsem);
-			ib_unregister_client(client);
-			return ret;
-		}
+		if (ret)
+			goto out;
 	}
+	ret = 0;
+out:
+	up_write(&clients_rwsem);
 	up_read(&devices_rwsem);
-	return 0;
+	if (need_unreg && ret)
+		ib_unregister_client(client);
+	return ret;
 }
 EXPORT_SYMBOL(ib_register_client);
Jason Gunthorpe Jan. 4, 2024, 12:37 p.m. UTC | #6
On Thu, Jan 04, 2024 at 02:48:14PM +0800, Shifeng Li wrote:

> The root cause is that mad_client and cm_client may init concurrently
> when devices_rwsem write semaphore is downgraded in enable_device_and_get() like:

That can't be true, the module loader infrastructue ensures those two
things are sequential.

You are trying to say that the post-client fixup stuff will still see
the DEVICE_REGISTERED before it reaches the clients_rwsem lock?

That probably just says the clients_rwsem should be obtained before
changing the DEVICE_STATE too :\

Jason
Shifeng Li Jan. 5, 2024, 8:15 a.m. UTC | #7
On 2024/1/4 20:37, Jason Gunthorpe wrote:
> On Thu, Jan 04, 2024 at 02:48:14PM +0800, Shifeng Li wrote:
> 
>> The root cause is that mad_client and cm_client may init concurrently
>> when devices_rwsem write semaphore is downgraded in enable_device_and_get() like:
> 
> That can't be true, the module loader infrastructue ensures those two
> things are sequential.
> 

I'm a bit confused how the module loader infrastructue ensures that mad_client.add() and
cm_client.add() are sequential. Could you explain in more detail please?

We know that the ib_cm driver and mlx5_ib driver can load concurrently.

Thanks.

> You are trying to say that the post-client fixup stuff will still see
> the DEVICE_REGISTERED before it reaches the clients_rwsem lock?
> 
> That probably just says the clients_rwsem should be obtained before
> changing the DEVICE_STATE too :\
> 
> Jason
>
Jason Gunthorpe Jan. 5, 2024, 2:19 p.m. UTC | #8
On Fri, Jan 05, 2024 at 04:15:18PM +0800, Shifeng Li wrote:
> On 2024/1/4 20:37, Jason Gunthorpe wrote:
> > On Thu, Jan 04, 2024 at 02:48:14PM +0800, Shifeng Li wrote:
> > 
> > > The root cause is that mad_client and cm_client may init concurrently
> > > when devices_rwsem write semaphore is downgraded in enable_device_and_get() like:
> > 
> > That can't be true, the module loader infrastructue ensures those two
> > things are sequential.
> > 
> 
> I'm a bit confused how the module loader infrastructue ensures that mad_client.add() and
> cm_client.add() are sequential. Could you explain in more detail
> please?

ib_cm has a symbol dependency on ib_mad, so the module loader will not
allow ib_cm to start running until all its symbol dependencies have
completed loading.

> We know that the ib_cm driver and mlx5_ib driver can load concurrently.

Yes, this is possible

Jason
Ding Hui Jan. 6, 2024, 2:12 a.m. UTC | #9
On 2024/1/4 20:37, Jason Gunthorpe wrote:
> On Thu, Jan 04, 2024 at 02:48:14PM +0800, Shifeng Li wrote:
> 
>> The root cause is that mad_client and cm_client may init concurrently
>> when devices_rwsem write semaphore is downgraded in enable_device_and_get() like:
> 
> That can't be true, the module loader infrastructue ensures those two
> things are sequential.
> 

Please consider the sequence again and notice that:

1. We agree that dependencies ensure mad_client be registered before cm_client.
2. But the mad_client.add() is not invoked in ib_register_client(), since
    there is no DEVICE_REGISTERED device at that time.
    Instead, it will be delayed until the device driver init (e.g. mlx5_core)
    in enable_device_and_get().
3. The ib_cm and mlx5_core can be loaded concurrently, after setting DEVICE_REGISTERED
    and downgrade_write(&devices_rwsem) in enable_device_and_get(), there is a chance
    that cm_client.add() can be invoked before mad_client.add().


         T1(ib_core init)      |      T2(device driver init)        |        T3(ib_cm init)
---------------------------------------------------------------------------------------------------
ib_register_client mad_client
   assign_client_id
     add clients CLIENT_REGISTERED
     (with clients_rwsem write)
   down_read(&devices_rwsem);
   xa_for_each_marked (&devices, DEVICE_REGISTERED)
     nop # no devices
   up_read(&devices_rwsem);

                               ib_register_device
                                 enable_device_and_get
                                   down_write(&devices_rwsem);
                                   set DEVICE_REGISTERED
                                   downgrade_write(&devices_rwsem);
                                                                     ib_register_client cm_client
                                                                       assign_client_id
                                                                         add clients CLIENT_REGISTERED
                                                                         (with clients_rwsem write)
                                                                       down_read(&devices_rwsem);
                                                                       xa_for_each_marked (&devices, DEVICE_REGISTERED)
                                                                         add_client_context
                                                                           down_write(&device->client_data_rwsem);
                                                                           get CLIENT_DATA_REGISTERED
                                                                           downgrade_write(&device->client_data_rwsem);
                                                                           cm_client.add
                                                                             cm_add_one
                                                                               ib_register_mad_agent
                                                                                 ib_get_mad_port
                                                                                   __ib_get_mad_port return NULL!
                                                                           set CLIENT_DATA_REGISTERED
                                                                           up_read(&device->client_data_rwsem);
                                                                       up_read(&devices_rwsem);
                                 down_read(&clients_rwsem);
                                 xa_for_each_marked (&clients, CLIENT_REGISTERED)
                                   add_client_context [mad]
                                     mad_client.add
                                   add_client_context [cm]
                                     nop # already CLIENT_DATA_REGISTERED
                                 up_read(&clients_rwsem);
                                 up_read(&devices_rwsem);

> You are trying to say that the post-client fixup stuff will still see
> the DEVICE_REGISTERED before it reaches the clients_rwsem lock?
> 
> That probably just says the clients_rwsem should be obtained before
> changing the DEVICE_STATE too :\
>
Shifeng Li Jan. 15, 2024, 3:27 a.m. UTC | #10
On 2024/1/5 22:19, Jason Gunthorpe wrote:
> On Fri, Jan 05, 2024 at 04:15:18PM +0800, Shifeng Li wrote:
>> On 2024/1/4 20:37, Jason Gunthorpe wrote:
>>> On Thu, Jan 04, 2024 at 02:48:14PM +0800, Shifeng Li wrote:
>>>
>>>> The root cause is that mad_client and cm_client may init concurrently
>>>> when devices_rwsem write semaphore is downgraded in enable_device_and_get() like:
>>>
>>> That can't be true, the module loader infrastructue ensures those two
>>> things are sequential.
>>>
>>
>> I'm a bit confused how the module loader infrastructue ensures that mad_client.add() and
>> cm_client.add() are sequential. Could you explain in more detail
>> please?
> 
> ib_cm has a symbol dependency on ib_mad, so the module loader will not
> allow ib_cm to start running until all its symbol dependencies have
> completed loading.

I have found a method to reproduce the issue as follow:

1. modprobe -r ib_cm; modprobe -r ib_core; modprobe -r mlx5_ib;
2. Compile and replace ib_core with following patch;
3. modprobe ib_cm;
4. modprobe mlx5_ib;

diff -Narup ./drivers/infiniband/core/device.c ./drivers/infiniband/core/device.c.reproduce
--- ./drivers/infiniband/core/device.c  2024-01-15 11:14:08.063094430 +0800
+++ ./drivers/infiniband/core/device.c.reproduce        2024-01-15 11:16:22.577096953 +0800
@@ -64,6 +64,8 @@ struct workqueue_struct *ib_wq;
  EXPORT_SYMBOL_GPL(ib_wq);
  static struct workqueue_struct *ib_unreg_wq;
  
+int ib_cm_run_flag = 0;
+
  /*
   * Each of the three rwsem locks (devices, clients, client_data) protects the
   * xarray of the same name. Specifically it allows the caller to assert that
@@ -1371,6 +1373,9 @@ static int enable_device_and_get(struct
          */
         downgrade_write(&devices_rwsem);
  
+       ib_cm_run_flag = 1;
+       msleep(30000);
+
         if (device->ops.enable_driver) {
                 ret = device->ops.enable_driver(device);
                 if (ret)
@@ -1843,6 +1848,12 @@ int ib_register_client(struct ib_client
         if (ret)
                 return ret;
  
+       if (strncmp(client->name, "cm", strlen("cm")) == 0) {
+               while (!ib_cm_run_flag) {
+                       cond_resched();
+               }
+       }
+
         down_read(&devices_rwsem);
         xa_for_each_marked (&devices, index, device, DEVICE_REGISTERED) {
                 ret = add_client_context(device, client);

> 
>> We know that the ib_cm driver and mlx5_ib driver can load concurrently.
> 
> Yes, this is possible
> 
> Jason
>
Jason Gunthorpe Jan. 15, 2024, 1:47 p.m. UTC | #11
On Sat, Jan 06, 2024 at 10:12:17AM +0800, Ding Hui wrote:
> On 2024/1/4 20:37, Jason Gunthorpe wrote:
> > On Thu, Jan 04, 2024 at 02:48:14PM +0800, Shifeng Li wrote:
> > 
> > > The root cause is that mad_client and cm_client may init concurrently
> > > when devices_rwsem write semaphore is downgraded in enable_device_and_get() like:
> > 
> > That can't be true, the module loader infrastructue ensures those two
> > things are sequential.
> > 
> 
> Please consider the sequence again and notice that:
> 
> 1. We agree that dependencies ensure mad_client be registered before cm_client.
> 2. But the mad_client.add() is not invoked in ib_register_client(), since
>    there is no DEVICE_REGISTERED device at that time.
>    Instead, it will be delayed until the device driver init (e.g. mlx5_core)
>    in enable_device_and_get().
> 3. The ib_cm and mlx5_core can be loaded concurrently, after setting DEVICE_REGISTERED
>    and downgrade_write(&devices_rwsem) in enable_device_and_get(), there is a chance
>    that cm_client.add() can be invoked before mad_client.add().
> 
> 
>         T1(ib_core init)      |      T2(device driver init)        |        T3(ib_cm init)
> ---------------------------------------------------------------------------------------------------
> ib_register_client mad_client
>   assign_client_id
>     add clients CLIENT_REGISTERED
>     (with clients_rwsem write)
>   down_read(&devices_rwsem);
>   xa_for_each_marked (&devices, DEVICE_REGISTERED)
>     nop # no devices
>   up_read(&devices_rwsem);
> 
>                               ib_register_device
>                                 enable_device_and_get
>                                   down_write(&devices_rwsem);
>                                   set DEVICE_REGISTERED
>                                   downgrade_write(&devices_rwsem);
>                                                                     ib_register_client cm_client
>                                                                       assign_client_id
>                                                                         add clients CLIENT_REGISTERED
>                                                                         (with clients_rwsem write)
>                                                                       down_read(&devices_rwsem);
>                                                                       xa_for_each_marked (&devices, DEVICE_REGISTERED)
>                                                                         add_client_context
>                                                                           down_write(&device->client_data_rwsem);
>                                                                           get CLIENT_DATA_REGISTERED
>                                                                           downgrade_write(&device->client_data_rwsem);
>                                                                           cm_client.add
>                                                                             cm_add_one
>                                                                               ib_register_mad_agent
>                                                                                 ib_get_mad_port
>                                                                                   __ib_get_mad_port return NULL!
>                                                                           set CLIENT_DATA_REGISTERED
>                                                                           up_read(&device->client_data_rwsem);
>                                                                       up_read(&devices_rwsem);
>                                 down_read(&clients_rwsem);
>                                 xa_for_each_marked (&clients, CLIENT_REGISTERED)
>                                   add_client_context [mad]
>                                     mad_client.add
>                                   add_client_context [cm]
>                                     nop # already CLIENT_DATA_REGISTERED
>                                 up_read(&clients_rwsem);
>                                 up_read(&devices_rwsem);

Take the draft I sent previously and use down_write(&devices_rwsem) in
ib_register_client()

Jason
Shifeng Li Jan. 26, 2024, 2:25 a.m. UTC | #12
On 2024/1/16 10:12, Ding Hui wrote:
> On 2024/1/15 21:47, Jason Gunthorpe wrote:
>> On Sat, Jan 06, 2024 at 10:12:17AM +0800, Ding Hui wrote:
>>> On 2024/1/4 20:37, Jason Gunthorpe wrote:
>>>> On Thu, Jan 04, 2024 at 02:48:14PM +0800, Shifeng Li wrote:
>>>>
>>>>> The root cause is that mad_client and cm_client may init concurrently
>>>>> when devices_rwsem write semaphore is downgraded in enable_device_and_get() like:
>>>>
>>>> That can't be true, the module loader infrastructue ensures those two
>>>> things are sequential.
>>>>
>>>
>>> Please consider the sequence again and notice that:
>>>
>>> 1. We agree that dependencies ensure mad_client be registered before cm_client.
>>> 2. But the mad_client.add() is not invoked in ib_register_client(), since
>>>     there is no DEVICE_REGISTERED device at that time.
>>>     Instead, it will be delayed until the device driver init (e.g. mlx5_core)
>>>     in enable_device_and_get().
>>> 3. The ib_cm and mlx5_core can be loaded concurrently, after setting DEVICE_REGISTERED
>>>     and downgrade_write(&devices_rwsem) in enable_device_and_get(), there is a chance
>>>     that cm_client.add() can be invoked before mad_client.add().
>>>
>>>
>>>          T1(ib_core init)      |      T2(device driver init)        |        T3(ib_cm init)
>>> ---------------------------------------------------------------------------------------------------
>>> ib_register_client mad_client
>>>    assign_client_id
>>>      add clients CLIENT_REGISTERED
>>>      (with clients_rwsem write)
>>>    down_read(&devices_rwsem);
>>>    xa_for_each_marked (&devices, DEVICE_REGISTERED)
>>>      nop # no devices
>>>    up_read(&devices_rwsem);
>>>
>>>                                ib_register_device
>>>                                  enable_device_and_get
>>>                                    down_write(&devices_rwsem);
>>>                                    set DEVICE_REGISTERED
>>>                                    downgrade_write(&devices_rwsem);
>>>                                                                      ib_register_client cm_client
>>>                                                                        assign_client_id
>>>                                                                          add clients CLIENT_REGISTERED
>>>                                                                          (with clients_rwsem write)
>>>                                                                        down_read(&devices_rwsem);
>>>                                                                        xa_for_each_marked (&devices, DEVICE_REGISTERED)
>>>                                                                          add_client_context
>>>                                                                            down_write(&device->client_data_rwsem);
>>>                                                                            get CLIENT_DATA_REGISTERED
>>>                                                                            downgrade_write(&device->client_data_rwsem);
>>>                                                                            cm_client.add
>>>                                                                              cm_add_one
>>>                                                                                ib_register_mad_agent
>>>                                                                                  ib_get_mad_port
>>>                                                                                    __ib_get_mad_port return NULL!
>>>                                                                            set CLIENT_DATA_REGISTERED
>>>                                                                            up_read(&device->client_data_rwsem);
>>>                                                                        up_read(&devices_rwsem);
>>>                                  down_read(&clients_rwsem);
>>>                                  xa_for_each_marked (&clients, CLIENT_REGISTERED)
>>>                                    add_client_context [mad]
>>>                                      mad_client.add
>>>                                    add_client_context [cm]
>>>                                      nop # already CLIENT_DATA_REGISTERED
>>>                                  up_read(&clients_rwsem);
>>>                                  up_read(&devices_rwsem);
>>
>> Take the draft I sent previously and use down_write(&devices_rwsem) in
>> ib_register_client()
>>
> 
> I believe this modification is effective, rather than expanding the clients_rwsem range,
> the key point is down_write(&devices_rwsem), which prevents ib_register_client() from
> being executed in the gap of ib_register_device().
> 
> However, this may cause a little confusion, as ib_register_client() does not modify
> anything related to devices, but it is protected by a write lock.
> 

Hi Jason,

Do you have any differing opinions about above?
Jason Gunthorpe Feb. 1, 2024, 12:04 a.m. UTC | #13
On Fri, Jan 26, 2024 at 10:25:01AM +0800, Shifeng Li wrote:
> > I believe this modification is effective, rather than expanding the clients_rwsem range,
> > the key point is down_write(&devices_rwsem), which prevents ib_register_client() from
> > being executed in the gap of ib_register_device().
> > 
> > However, this may cause a little confusion, as ib_register_client() does not modify
> > anything related to devices, but it is protected by a write lock.
> > 
> Do you have any differing opinions about above?

I'd rather see the client side have a comment explaining the confusing
lock then disturb the locking on the devices side..

I think the write/read lock scheme there was done for a reason I just
haven't had time to recall exactly why..

Please send a patch with this arrangement

Jason
diff mbox series

Patch

diff --git a/drivers/infiniband/core/device.c b/drivers/infiniband/core/device.c
index 67bcea7a153c..85782786993d 100644
--- a/drivers/infiniband/core/device.c
+++ b/drivers/infiniband/core/device.c
@@ -1315,12 +1315,6 @@  static int enable_device_and_get(struct ib_device *device)
 	down_write(&devices_rwsem);
 	xa_set_mark(&devices, device->index, DEVICE_REGISTERED);
 
-	/*
-	 * By using downgrade_write() we ensure that no other thread can clear
-	 * DEVICE_REGISTERED while we are completing the client setup.
-	 */
-	downgrade_write(&devices_rwsem);
-
 	if (device->ops.enable_driver) {
 		ret = device->ops.enable_driver(device);
 		if (ret)
@@ -1337,7 +1331,7 @@  static int enable_device_and_get(struct ib_device *device)
 	if (!ret)
 		ret = add_compat_devs(device);
 out:
-	up_read(&devices_rwsem);
+	up_write(&devices_rwsem);
 	return ret;
 }